On Wed, Jun 06, 2018 at 08:17:54PM +0200, Remi Locherer wrote:
> On Wed, Jun 06, 2018 at 09:01:49AM +0200, Claudio Jeker wrote:
> > On Wed, Jun 06, 2018 at 08:06:30AM +0200, Remi Locherer wrote:
> > > Hi,
> > >
> > > RfC 5340 says that for intra area prefix LSAs metric should be set to 0
> > > in case of point-to-multipoint or loopback interfaces. Otherwise metric
> > > should be set to the value of the interfaces output cost.
> > >
> > > ospf6d currently sends intra area prefix LSAs *always* with metric 0.
> > >
> > > Below diff fixes this.
> > >
> > > OK?
> > >
> > > Remi
> > >
> > >
> > > Index: rde.c
> > > ===================================================================
> > > RCS file: /cvs/src/usr.sbin/ospf6d/rde.c,v
> > > retrieving revision 1.72
> > > diff -u -p -r1.72 rde.c
> > > --- rde.c 12 Aug 2017 16:27:50 -0000 1.72
> > > +++ rde.c 5 Jun 2018 20:45:03 -0000
> > > @@ -1301,7 +1301,6 @@ append_prefix_lsa(struct lsa **lsa, u_in
> > > /* Append prefix to LSA. */
> > > copy = (struct lsa_prefix *)(new_lsa + *len);
> > > memcpy(copy, prefix, lsa_prefix_len);
> > > - copy->metric = 0;
> > >
> > > *lsa = (struct lsa *)new_lsa;
> > > *len = new_len;
> > > @@ -1514,6 +1513,7 @@ orig_intra_lsa_rtr(struct area *area, st
> > > if (iface->type == IF_TYPE_POINTOMULTIPOINT ||
> > > iface->state & IF_STA_LOOPBACK) {
> > > lsa_prefix->prefixlen = 128;
> > > + lsa_prefix->metric = 0;
> >
> > This is not strictly needed because there is a bzero done before.
>
> I thought its easier to spot that metric is 0 on purpose.
> Same in the updated diff below in prefix_tree_add (calloc()).
>
> What is more common in such a case?
>
> >
> > > } else {
> > > lsa_prefix->prefixlen = ia->prefixlen;
> > > lsa_prefix->metric = htons(iface->metric);
> > >
> >
> > What about the append_prefix_lsa() in orig_intra_lsa_net()?
> > Currently the metric is set to 0 there but now it wont.
> > What should the behaviour be there?
>
> Metric should be set to 0.
>
> > I guess it would be possible to change prefix_tree_add() to set metric to
> > 0 there. Also I think prefix_tree_add() is leaking memory.
> > The if (!(IN6_IS_ADDR_LINKLOCAL(&addr)) && ... should have an
> > else free(new); block to make sure we don't leak new if it is not added to
> > the prefix_tree.
>
>
> Index: rde.c
> ===================================================================
> RCS file: /cvs/src/usr.sbin/ospf6d/rde.c,v
> retrieving revision 1.72
> diff -u -p -r1.72 rde.c
> --- rde.c 12 Aug 2017 16:27:50 -0000 1.72
> +++ rde.c 6 Jun 2018 16:29:46 -0000
> @@ -1301,7 +1301,6 @@ append_prefix_lsa(struct lsa **lsa, u_in
> /* Append prefix to LSA. */
> copy = (struct lsa_prefix *)(new_lsa + *len);
> memcpy(copy, prefix, lsa_prefix_len);
> - copy->metric = 0;
>
> *lsa = (struct lsa *)new_lsa;
> *len = new_len;
> @@ -1354,6 +1353,8 @@ prefix_tree_add(struct prefix_tree *tree
> memcpy(&addr, new->prefix + 1,
> LSA_PREFIXSIZE(new->prefix->prefixlen));
>
> + new->prefix->metric = 0;
> +
> if (!(IN6_IS_ADDR_LINKLOCAL(&addr)) &&
> (new->prefix->options & OSPF_PREFIX_NU) == 0 &&
> (new->prefix->options & OSPF_PREFIX_LA) == 0) {
> @@ -1362,7 +1363,8 @@ prefix_tree_add(struct prefix_tree *tree
> old->prefix->options |= new->prefix->options;
> free(new);
> }
> - }
> + } else
> + free(new);
>
> cur_prefix = cur_prefix + len;
> }
> @@ -1514,6 +1516,7 @@ orig_intra_lsa_rtr(struct area *area, st
> if (iface->type == IF_TYPE_POINTOMULTIPOINT ||
> iface->state & IF_STA_LOOPBACK) {
> lsa_prefix->prefixlen = 128;
> + lsa_prefix->metric = 0;
> } else {
> lsa_prefix->prefixlen = ia->prefixlen;
> lsa_prefix->metric = htons(iface->metric);
>
Looks good. Maybe split the memleak fix from the metric bits and commit
that independently. OK claudio@
--
:wq Claudio
